I'm Kyle, a Principal Applied Science Lead specializing in Responsible AI and LLM evaluation systems. At Microsoft, I lead efforts to develop LLM-based safety evaluations and automated red-teaming solutions.
My recent work focuses on responsible AI, building systems to detect and prevent harmful content generation while advancing language model capabilities.
What I'm Working On
At Microsoft, I lead a team of research scientists addressing key challenges in Responsible AI:
I develop "LLM-Judges" that automatically evaluate potentially harmful content across Microsoft's AI products. This work was published in A Framework for Automated Measurement of Responsible AI Harms in Generative AI Applications.
My team creates automated red-teaming and synthetic data generation platforms to proactively identify risks in generative models before deployment.
Team Publications
Recent publications from my team where I contributed in an advisory capacity:
- Zhang, R., Sullivan, D., et al. "Defense against Prompt Injection Attacks via Mixture of Encodings" (NAACL 2025) - Research on prompt injection mitigations led by my team.
- Zhang, J., Elgohary, A., et al. "Controllable Safety Alignment: Inference-Time Adaptation to Diverse Safety Requirements" - Research on adapting language models to different safety requirements at inference time.